基于2d-距离改进的K-means聚类算法研究

为了解决原始K-means算法随机选取聚类中心对聚类结果产生较大影响的不足和孤立点的存在对聚类精度的破坏,以及两者之间的相互牵制性,采用基于2d-距离的DKC值来对原始样本数据集进行预处理以分辨孤立点,同时确定初始的聚类中心,达到消除两者相互影响的效果,使得聚类中心相对稳定,改进后的算法较原始的算法在准确度上得到了改进。

Research on improved K-means Clustering Algorithm Based on Two-distance

In order to overcome the shortcoming of original K-means algorithm that randomly selecting clustering center puts much influence on clustering results,prevent the destruction on clustering precision resulted from the existance of isolated points,and reveal the inter relationship between them,this paper adopted DKC value of 2d-distance to pretreat original sample data to distingwish isolated points and determine initial clustering center,so as to eliminater the interrelationship and stablize clustering center.Compared with original algorithm,the improved one is more effective in accuracy.